Improve UPLC CSE performance.

During [the investigation](https://github.com/IntersectMBO/plutus-private/issues/1425) of slow compilation of RSNARKs prototype project CSE was identified as a bottleneck.

Here are some ideas on how to improve its performance:

- The CSE transformation in UPLC [operates on an approximate builtin function arity information](https://github.com/IntersectMBO/plutus/blob/master/plutus-core/untyped-plutus-core/src/UntypedPlutusCore/Transform/Cse.hs#L44-L47). It makes a [separate pass collecting builtin arities](https://github.com/IntersectMBO/plutus/blob/master/plutus-core/untyped-plutus-core/src/UntypedPlutusCore/Transform/Cse.hs#L63-L64). The idea is to avoid this pass relying on statically known builtin function arity information instead.
- Improve `isAncestorOrSelf` (uses `isSuffixOf` currently)
- Avoiding Unnecessary Traversals: The algorithm traverses the AST multiple times, and some operations may be redundant. Use a single pass transformation instead of multiple traversals.